[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH RFC 0/2] s390x: Make the s390-ccw BIOS relocatab
Re: [Qemu-devel] [PATCH RFC 0/2] s390x: Make the s390-ccw BIOS relocatable
Thu, 05 Mar 2015 13:49:28 +0100
Mozilla/5.0 (Macintosh; Intel Mac OS X 10.10; rv:31.0) Gecko/20100101 Thunderbird/31.5.0
On 05.03.15 13:43, Christian Borntraeger wrote:
> Am 05.03.2015 um 12:10 schrieb Alexander Graf:
>> On 05.03.15 12:02, Thomas Huth wrote:
>>> Currently, our s390-ccw.img sits at the fix address 126 MiB in memory.
>>> This has two big disadvantages: 1) We can not start guests with less
>>> than 128 MiB RAM and 2) if the guest uses a really huge ramdisk > 126 MiB,
>>> the s390-ccw BIOS gets overwritten and the boot silently crashes.
>>> These two patches now fix these problems by relocating the s390-ccw
>>> BIOS to the end of the RAM. The basic idea here is to compile the
>>> BIOS with the "-fpie" compiler option to create position independent
>>> code. Sounds easy at a first glance - however, with -fpie, we only
>>> get position independent _code_ - and a so called GOT (global offset
>>> table) which contains absolute references to global variables again
>>> (this is normally needed for supporting dynamic libraries - in our
>>> s390-ccw BIOS, it's just bad luck that we get a GOT).
>>> So to be able to really move around our s390-ccw.img in RAM, we've got
>>> to relocate the entries in the GOT, too. This is what the first patch
>>> is good for. I've changed the ELF loader there to parse the reloc section
>>> of the ELF file. I only included the bare minimum of relocation types
>>> (R_390_RELATIVE) in the patch so far, but this can easily be extended
>>> in case we need more (with different compiler version etc.).
>>> The second patch then adds the required changes to the s390-ccw BIOS
>>> Makefile and our s390 ipl code in QEMU.
>>> Now I'd like to get some feedback on this approach: Is it ok to extend
>>> the ELF loader this way? Does anybody have better/nicer ideas to solve
>>> the problem of a relocatable BIOS?
>>> Thanks for any insights!
>> I think the approach is perfectly valid and good :)
> Yes, looks good. It relocated much higher, but not beyond 2GB, and it can
> now handle guests < 128MB.
> I can take these patches via the s390 tree.
> Would be good if somebody else (Paolo,Alex ??)
> could Ack the change in include/hw/elf_ops.h.
Acked-by: Alexander Graf <address@hidden>